Research On On-Line Curvature Detection Technology Of Profile Roll Bend

On-line dimension detection of aluminum alloy profile rolling with variable curvature is the key to realize closed-loop control of rolling.In view of the lack of effective on-line dimension detection method during rolling with variable curvature of aluminum alloy profile,this paper proposes an on-line dimension detection method based on follow-up machine vision.The camera moves synchronously with the bending roller,and shoots the area after the springback of the profile rolling.The photographed images are spliced by using the bidirectional uniqueness matching SURF algorithm,and the outer edge curve of the profile rolling part is extracted.On this basis,the curve size is detected and matched with the digital-analog curve,and the error distribution of the profile rolling part is obtained,which meets the detection requirements of the variable curvature rolling forming size of aluminum alloy profiles.The main work of present paper was shown:(1)Research on sectional splicing algorithm of profile rolling bending.By adding external image features and based on the profile rolling characteristics,the range budget of image feature area is realized.Image feature matching is realized by SURF feature matching algorithm combined with bidirectional unique matching.By calculating the angle between the images,the self-adaptation of canvas allocation space of the rolled mosaic image is realized,and a large number of invalid areas are removed from the mosaic image.The gradual-in and gradual-out fusion method is improved to realize rolling image fusion;(2)The realization of local detection and overall detection of profile roll bending.Hough transform detection method is used to remove the remaining straight edges of profile rolled parts in the image,and slope difference adaptive method is used to sample image edge data points,which can be distributed with the curvature change of rolled parts.The problem of image edge missing is solved by the method of curve piecewise fitting,and the online unilateral comparison between the outer edge curve of the roll bending part and the roll bending numerical model is made,and the unilateral error is obtained.The particle swarm optimization algorithm based on simulated annealing realizes the matching between the measured edge curve and the digital-analog curve,and obtains the error distribution data of profile rolling parts;(3)Construction of curvature detection system for profile rolling image.The system design and functional module division of the detection system are carried out.Completed the selection of light source,camera position determination and camera calibration of the detection system,and developed the curvature detection software of profile rolling image;(4)On-line measurement experiment of profile roll bending.On-line measurement is carried out on several roll-bending parts with different numerical models,and then re-inspection is carried out by using 3D laser scanner.The re-inspection results verify the accuracy of the detection method in this paper.Finally,the same profile roll-bending part is tested several times,and the repeated error of the detection system in this paper is obtained.